Beyond ‘Best Practice’ – A Data-Driven Framework for Smarter Campaigns

For anyone managing a large e-commerce account, Performance Max presents a paradox. It promises unparalleled reach and efficiency, yet it can often feel like you’ve handed over the keys to a black box. You see the budget being spent, but on what? Why do the same handful of products soak up all the attention while thousands of others gather digital dust?

This isn’t a bug in the system; it’s a feature. 

And it was the exact challenge facing one of our clients, a major retailer with an inventory of over 335,000 products. Their move to PMax had led to a frustrating decline in performance. Despite their best efforts to segment campaigns by category, price, and ROAS, they were grappling with muddled attribution and felt they were losing their strategic grip.

Our mission was to move beyond the standard playbook, decode the machine’s behaviour, and align its power with what truly drove the client’s bottom line.

Diagnosing the Disconnect

Automated systems are designed to follow patterns, and at scale, this can lead to outcomes that require intervention:

• The first is a challenge of performance momentum. PMax is effective at identifying products with recent sales data and allocating more budget to them. This creates a momentum effect that, while efficient, can leave a significant portion of the product catalogue under-explored, including items with high potential but no recent performance history.
• Then you have a problem of blended averages. When products are grouped by general category, high-margin bestsellers are often mixed with low-margin clearance items. This forces the algorithm to bid based on an average performance that doesn’t reflect the true value of any individual product within the group.
• The last, and most important, you have a context gap. By default, the platform optimises for a conversion based on the data it has. It lacks the crucial business context to differentiate between a low-profit sale and a strategically valuable one, treating both as positive signals.

A workshop with the client’s marketing, commercial, and analytics teams confirmed the core issue: the campaign’s definition of “success” was not aligned with the business’s. 

The solution wasn’t to restrict the algorithm, but to provide it with a richer, more accurate set of data signals.

Building Intelligence into the Strategy

We developed a proprietary classification engine, to translate the client’s deep business knowledge into a clear signal that PMax could act on. This intelligence layer was built on four pillars:

1. How much demand potential was there? We analysed data from across the client’s ecosystem, including organic search and direct traffic, to forecast each product’s underlying demand. This allowed us to identify popular products that were being overlooked by the ad platform.
2. What was the economical performance of the product? We moved beyond tracking revenue to model for profitability. Using probabilistic machine learning, we analysed factors like CPC, conversion rates, and profit margins for each product to determine the viability of investing in it efficiently.
3. Was it the right time of year for the product? Using advanced statistical analysis, we identified products with reliable seasonal demand patterns. This enabled proactive budget allocation for major holidays as well as for emerging trends.
4. What was a unique signal for the client? We integrated the client’s unique operational data – including product ratings, stock levels, add-to-basket frequency, and delivery times – to create a comprehensive view of each product’s true commercial value.

Developing the Framework

This classification engine produced a five-tier segmentation system that sorted every product by its strategic role:

• Champions: Consistently high-performing products in terms of volume and profit.
• Challengers: Solid performers with clear potential for growth.
• Seeds: New or low-data items showing early promise, designated for controlled testing.
• Seasonal Stars: Products identified for activation during specific periods.
• Dormants: Low-potential items, removed from paid spend to improve efficiency.

Critically, this was a dynamic system. The product feed was re-analysed and re-labelled daily, allowing products to move between these tiers and their corresponding campaigns as their performance data changed.

The Final Step

With this framework in place, we implemented the final enhancement: we changed the campaign’s optimisation target from revenue to actual profit. By importing offline data on the commission earned from each sale, we gave PMax a clear directive that was perfectly aligned with the client’s financial goals.

The Result

The impact of this new approach was immediate and significant.

Over a 12-month evaluation period, the framework generated millions of pounds in incremental commission revenue – without increasing the overall advertising spend.

The key takeaway is that an advertiser’s first-party data is their most effective tool for managing automation. The path to success with PMax at scale is not about trying to outmanoeuvre the algorithm, but about enriching it. When you provide the system with a clear definition of what matters to your business, you create a powerful signal that drives genuinely meaningful growth.

Inderpaul Rai is Director, Paid Media at WeDiscover. This article is based on the talk he gave at Hero Conf UK in April 2025. Watch the recording in full here.